Add Cerebras adapter integration with custom streaming support

[AlexMikhalev]

Summary

This PR adds complete Cerebras Inference API integration to rust-genai with custom streaming support.

What's Included

• ✅ Custom Cerebras Streamer - Handles Cerebras's streaming format that ends with StreamEnded instead of [DONE]
• ✅ Full Chat Support - OpenAI-compatible chat completions with all standard parameters
• ✅ Streaming Chat - Real-time streaming responses with proper error handling
• ✅ JSON Mode - Structured output support
• ✅ Model Namespacing - cerebras::model_name format support
• ✅ Comprehensive Tests - Live provider tests for all supported features
• ✅ Example Code - Working example demonstrating usage
• ✅ CI Integration - Dedicated workflow for Cerebras tests

Key Features

• Models: cerebras::llama-3.1-8b, cerebras::llama-3.3-70b, and others
• Authentication: CEREBRAS_API_KEY environment variable
• Endpoint: https://api.cerebras.ai/v1/
• Streaming: Custom implementation handling Cerebras's stream termination
• Rate Limiting: Aware of ~3 requests/second limit

Files Added/Modified

• src/adapter/adapters/cerebras/streamer.rs - New custom streamer
• src/adapter/adapters/cerebras/adapter_impl.rs - Updated to use custom streamer
• src/adapter/adapters/cerebras/mod.rs - Added streamer module
• .github/workflows/cerebras-tests.yml - Dedicated CI workflow
• .github/workflows/ci.yml - Updated to compile Cerebras tests
• examples/c11-cerebras.rs - Usage example (already added)
• tests/tests_p_cerebras.rs - Live tests (already added)

Testing

All tests pass when run with API key:

• Chat completions ✅
• Streaming chat ✅
• JSON mode ✅
• Temperature control ✅
• Stop sequences ✅
• Model listing ✅
• Multi-system messages ✅

CI/CD

• Main CI compiles Cerebras code without API key
• Dedicated workflow runs live tests when CEREBRAS_API_KEY is configured
• Workflow triggers on changes to Cerebras-related files

Usage

let client = Client::default();
let model = "cerebras::llama-3.1-8b";
let response = client.exec_chat(model, chat_req, None).await?;

The integration is production-ready and follows all existing patterns in the codebase.